Narrow Results

In this paper, the physics-informed neural networks (PINNs) are applied to high-dimensional system to solve the $(N+1)$-dimensional initial-boundary value problem with $2N+1$ hyperplane boundaries. This method is used to solve the most classic (2+1)-dimensional integrable Kadomtsev–Petviashvili (KP) equation and (3+1)-dimensional reduced KP equation. The dynamics of (2+1)-dimensional local waves such as solitons, breathers, lump and resonance rogue are reproduced. Numerical results display that the magnitude of the error is much smaller than the wave height itself, so it is considered that the classical solutions in these integrable systems are well obtained based on the data-driven mechanism.

A new approach for solving computational mechanics problems using physics-informed neural networks (PINNs) is proposed. Variational forms of residuals for the governing equations of solid mechanics are utilized, and the residual is evaluated over the entire computational domain by employing domain decomposition and polynomials test functions. A parameter network is introduced and initial and boundary conditions, as well as data mismatch, are incorporated into a total loss function using a weighted summation. The accuracy of the model in solving forward problems of solid mechanics is demonstrated to be higher than that of the finite element method (FEM). Furthermore, homogeneous and heterogeneous material distributions can be effectively captured by the model using limited observations, such as strain components. This contribution is significant for potential applications in non-destructive evaluation, where obtaining detailed information about the material properties is difficult.